Incubator-mxnet: MXnet with cuda wont install on windows 10!

Created on 12 Jun 2020  路  5Comments  路  Source: apache/incubator-mxnet

Is gpu really supported in windows ?

tried using 1.6.0,1.5.0 with python 3.6,3.7,3.8 with cuda installed (verified using nvcc --version) and no matter what it wont impor t!

Cant find dlls, or OSError: [WinError 126] The specified module could not be found or new everytime!!

Bug Installation Windows

Most helpful comment

My MXNet with win10 is OK(although quite slow compared to Linux)
have you ever tried calling nvidia-smi?

Yes, i call it every day -l 1

If nvidia-smi is callable, the CUDA should be installed properly.
(you could check the "CUDA Version") slot to ensure it.

neutron@Neutron:/me$ nvidia-smi
Tue Jun 16 16:29:00 2020       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.82       Driver Version: 440.82       CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1060    On   | 00000000:01:00.0 Off |                  N/A |
| N/A   80C    P2    78W /  N/A |   2270MiB /  6078MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1003      G   /usr/lib/Xorg                                 38MiB |
|    0     96256      C   python                                      2227MiB |
+-----------------------------------------------------------------------------+

If the CUDA is installed properly, I don't know what would cause MXNet failed to load.
maybe pdb could help telling you what happened.

import pdb
pdb.set_trace()
import mxnet

save the following script into a test.py then execute python test.py, the python would enter pdb mode, then, using either s(step), n(next) or r(execute until return) to control the pdb procedure.
It may help you to find which dll is missing. and you may manually edit the .py script of MXNet to ensure a successful load.

All 5 comments

My MXNet with win10 is OK(although quite slow compared to Linux)
have you ever tried calling nvidia-smi?

My MXNet with win10 is OK(although quite slow compared to Linux)
have you ever tried calling nvidia-smi?

Yes, i call it every day -l 1

My MXNet with win10 is OK(although quite slow compared to Linux)
have you ever tried calling nvidia-smi?

Yes, i call it every day -l 1

If nvidia-smi is callable, the CUDA should be installed properly.
(you could check the "CUDA Version") slot to ensure it.

neutron@Neutron:/me$ nvidia-smi
Tue Jun 16 16:29:00 2020       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.82       Driver Version: 440.82       CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1060    On   | 00000000:01:00.0 Off |                  N/A |
| N/A   80C    P2    78W /  N/A |   2270MiB /  6078MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1003      G   /usr/lib/Xorg                                 38MiB |
|    0     96256      C   python                                      2227MiB |
+-----------------------------------------------------------------------------+

If the CUDA is installed properly, I don't know what would cause MXNet failed to load.
maybe pdb could help telling you what happened.

import pdb
pdb.set_trace()
import mxnet

save the following script into a test.py then execute python test.py, the python would enter pdb mode, then, using either s(step), n(next) or r(execute until return) to control the pdb procedure.
It may help you to find which dll is missing. and you may manually edit the .py script of MXNet to ensure a successful load.

@mxnet-label-bot add [windows]

[WinError 126]Because you lost some DLL dependencies.use https://github.com/lucasg/Dependencies/releases check depend with mxnet_xx.dll

Was this page helpful?
0 / 5 - 0 ratings